Robotic manipulators are seen as a potentially important instrument for rehabilitation because they can be programmed to exert a variety of forces (force fields) while the subject moves. They can also be used repetitively, while monitoring and recording subject's performance, allowing more intensive and prolonged treatment than is currently feasible with a human therapist. The question that remains is what type of force field is best suited for assisting in the recovery of motor function in hemiparetic human subjects, and how should the stroke survivor be trained in such a force field. When normal subjects are exposed to a novel force field that disturbs their motion in a systematic fashion, they adapt progressively so that they exhibit characteristic after-effects when the disturbing force field is unexpectedly removed. Both the initial adaptation and the after-effects occur without subjects being aware of the adaptive process. Since after-effects can be predicted by dynamic models of the neuro-musculoskeletal system, I plan to use dynamic modeling techniques as a tool for designing forces that will yield desirable after- effects in hemiparetic stroke patients that show stereotypical errors in movement. Currently there is evidence that hemiparetic stroke patients recovery is enhanced when they experience assistive force fields. Other evidence suggests that training against resistive force fields during reaching movements may also help the patient perform a desirable motion once the resistive field is removed. The main goal of this project is to determine which type of field -- assistive or resistive -- is best suited to attaining desirable rehabilitation outcomes. To this end I will use a programmable two-link robot that exerts forces during reaching movements in the horizontal plane. I plan to determine first to what extent after-affects can be made to persist. I will then develop and test a computational framework that designs assistive and resistive force fields. Finally will determine whether assistive or resistive force fields are best suited for mediating the desired compensatory effects on movements in hemiparetic stroke survivors. The results of these experiments should favor the development and application of more sophisticated rehabilitation environments using robots and provide a clear direction for more long-term clinical trials in the area of robot assisted rehabilitation.